• Aucun résultat trouvé

L’INSTITUTFOURIER DE ANNALES

N/A
N/A
Protected

Academic year: 2022

Partager "L’INSTITUTFOURIER DE ANNALES"

Copied!
34
0
0

Texte intégral

(1)

L’INSTITUT FOURIER

David HARARI & Alexei N. SKOROBOGATOV

The Brauer group of torsors and its arithmetic applications Tome 53, no7 (2003), p. 1987-2019.

<http://aif.cedram.org/item?id=AIF_2003__53_7_1987_0>

© Association des Annales de l’institut Fourier, 2003, tous droits réservés.

L’accès aux articles de la revue « Annales de l’institut Fourier » (http://aif.cedram.org/), implique l’accord avec les conditions générales d’utilisation (http://aif.cedram.org/legal/). Toute re- production en tout ou partie cet article sous quelque forme que ce soit pour tout usage autre que l’utilisation à fin strictement per- sonnelle du copiste est constitutive d’une infraction pénale. Toute copie ou impression de ce fichier doit contenir la présente mention de copyright.

cedram

Article mis en ligne dans le cadre du

Centre de diffusion des revues académiques de mathématiques http://www.cedram.org/

(2)

THE BRAUER GROUP OF TORSORS AND ITS ARITHMETIC APPLICATIONS

by

D. HARARI and A. N. SKOROBOGATOV

Introduction.

This paper consists of two

parts.

In the first

part

we describe the behaviour of the Brauer group with

respect

to the

pull-back

map from the

variety

to a torsor under a torus over any field of characteristic 0.

To a

variety

X endowed with a

surjective morphism

: X ---+

P~

with

geometrically integral generic

fibre one

canonically

associates certain X-

torsors,

called vertical torsors. The second

part

of the paper concerns the arithmetic of vertical torsors over a number field k. Let K =

k(t)

be the

function field of

P’,

and let K be an

algebraic

closure of K. Assume that the Picard group of the

geometric generic

fibre

XK

is

finitely generated,

and

that its Brauer group is finite. These conditions are

satisfied,

for

example,

when

XK

is

rationally

connected. We prove that the Manin obstruction to the Hasse

principle

and weak

approximation

on smooth and proper models of such torsors is the

only

one

provided

the same

property

holds for the k-fibres of 7r, and the number of ’bad’ fibres is small.

Using

the results of the first

part

we

give

sufficient conditions for X itself

(strictly speaking,

for

any smooth and proper model of

X)

to have the afore-mentioned

property.

To illustrate

possible applications

we exhibit

apparently

new classes of conic bundle threefolds and some other varieties with the

property

that the

Keywords: Brauer group - Hasse principle - Universal torsor.

Math. classification: llG35 - 14G05.

(3)

f :

Y - X

variety

X such that all invertible functions on X are

constants,

and Pic X has no

torsion, they

prove that the kernel of the natural map Br Y - Br Y is

naturally isomorphic

to Br k. In this paper

we show

(Theorem 1.7)

that the map Br X -

Br Y/Br k

defined

by f

is

canonically

identified with the map Br X -

(Br X )r (provided

k is such

that

H3 (r,1~* )

=

0).

We also obtain sufficient conditions that ensure that the unramified Brauer group of Y is the

image

of the unramified Brauer group of X.

In the arithmetic

part

of the paper we

present

a

generalization

of the

results of

[H94],

as well as of those of

[S90], [S96], [CSOO],

that allows a

few bad fibres in the theorem on the behaviour of the Manin obstruction

regarding

the passage from the k-fibres to the total space. Our main achievement are Theorems 2.12 and

2.9, proved by

a combination of

’open

descent’ and fibration methods. Let X be a smooth and

projective variety

over a number field

k,

and let X -

P~

be a dominant

morphism

with

geometrically integral generic

fibre. Assume that the sum of

degrees

of the

closed

points

of

Pk corresponding

to

non-split (bad)

fibres is at most

3,

and

that every

non-split

fibre contains a

multiplicity

1 irreducible

component splitting

into two over an

algebraic

closure of the residue field. Assume further that

PicXK

is

finitely generated, Br XK

is

finite,

and

XK

has a

k(t)-point.

Then we prove that unless we are in an

explicitly

described

exceptional

case, the

possible

failure of the Hasse

principle

and weak

approximation

on X is accounted for

by

the Manin obstruction whenever the same is true for smooth k-fibres of 1T. We also prove a similar result in the case when there are

only

two

non-split fibres,

both over

k-points,

each

containing

an irreducible

component

of

multiplicity

1 defined over an

extension of k of

prime degree. Explicit examples

of

applications

can be

found in the end of the paper.

(4)

Notation and conventions

All

cohomology

groups in this paper are either group

cohomology

of

finite groups, Galois or 6tale

cohomology

groups.

(For example,

we write

7~(~,...)

for

H6t(X,...)

for a scheme

X.)

Let F be a finite group, and M be an F-module. Then

M)

denotes the

subgroup

of

M)

con-

sisting

of the classes whose

image

under the restriction map to every

cyclic subgroup

of F is zero. For a discrete r-module M we write

M)

for the

Galois

cohomology

group

Hi (r, M).

Assume that

M,

as an abelian group, is of finite

type

and torsion-free. We define for i =

1, 2,

where G is the

(finite) image

of the action of r on M. Note that the groups

M),

i =

1, 2,

are finite. Let 1~’ be the smallest extension of k such that the action of Gal

(k/k’)

on M is trivial. Write G = Gal

(k’lk).

Then

H’(k’, M)

= 0 because

HI (k’, Z)

= 0. The restriction-inflation se-

quence for M

gives M) = Hl (G, M).

There is also an exact sequence

We conclude that

M)

is

naturally

a

subgroup

of

M).

When 1~ is a number field we denote

by Qk

the set of

places

of

k,

and

by kv

the

completion

of k at the

place

v. Let

Ak

be the

ring

of adeles of k. It is an easy

corollary

of the Tchebotarev

density

theorem that for a

number field k the group

M)

consists of a E

M)

such that

the localisation cxv E

M)

is trivial for almost all

places

v E

Ok.

By

a

variety

X over a field k we understand in this paper a

separated

scheme of finite

type

over

Speck.

The Brauer group Br X =

H2 (X, Gm)

is

equipped

with a natural filtration

Bro X

C

Br1 X

C

Br X,

where

BrX]

and Set

A;[X]*

=

HO(X, Cm ) .

There is a Hochschild-Serre

spectral

sequence

J

It

yields

a canonical map When we

obtain the exact sequence of low

degree

terms

and a

complex

(5)

Recall that for any number field k the local invariant of class field

theory

is an

injective map jv : Br kv

-~ which is an

isomorphism

for

finite v. Let X be a proper, smooth and

geometrically integral k-variety.

Then the set of adelic

points X(Ak)

is the

product TIvEOk

Assume

that

0,

and let

X(k)

be the closure of the

diagonal image

of

X(k)

in

X(Ak ) .

Set

(Note

that the sum is well

defined,

see

[CS87a]

III or

[SO1], 5.2).

The

reciprocity

law of

global

class field

theory implies

that

In

particular,

the condition

0

is an obstruction to the existence of a k-rational

point

on

X;

this is the Manin

(or Brauer-Manin)

obstruction to the Hasse

principle.

The condition

X (Ak)

is the Brauer- Manin obstruction to weak

approximation.

One says that the Brauer- Manin obstruction to weak

approximation

is the

only

one if

X(k) =

The

property

that the Brauer-Manin obstruction to the Hasse

principle (resp.

to weak

approximation)

is the

only

one is a k-birational invariant of smooth and proper varieties.

(This

follows from the birational invariance of the Brauer group

([G], 6.2),

the v-adic

implicit

function

theorem,

and Nishimura’s

lemma,

which says that the condition

X (1~) ~ ~

is a k-birational invariant of smooth and proper

varieties.)

1. Torsors

and the

Brauer group.

In this section k is any field of characteristic zero.

(6)

1.1. Torsors under tori and their relative

compactifications.

By

T we shall denote an

algebraic

k-torus with module of characters

T.

Let p : X 2013~

Speck

be a smooth and

geometrically integral k-variety,

and let

f :

V 2013~ X be a torsor under T.

Equivalently,

Y is a smooth and

geometrically integral k-variety equipped

with a

(scheme-theoretically)

free

action of T such that

YIT

= X.

We shall

frequently

use the

Leray spectral

sequence

By

the relative Rosenlicht lemma

([CS87a], Prop. 1.4.2)

we have an exact

sequence of sheaves on X in the 6tale

topology:

Here

p*T

is a

sheaf

on X obtained from the F-module

T.

Note that

Z)

= 0 for any normal

integral

scheme S

([SGA 4],

IX 3.6

(ii)j.

This

implies

in

particular

that = 0.

LEMMA 1.1. - Let X be a normal

k-variety,

and let

f :

Y ---+ X be a

torsor under a torus T. Then we have the

following properties:

Proof.

(i)

is established in the

proof

of

([CS87a], Prop. 2.1.1 ) .

For

the convenience of the reader we

reproduce

this

argument

here in the case T =

G, alongside

with the

proof

of

(ii).

Let x be a

geometric point

of

X,

R be the

strictly

henselian local

ring

of X at x, and K be the field of fractions of R. The stalk of at

x is

Pic (Y

x x

R),

and the stalk of

R2 f*Gm

at x is Br

(Y

Xx

R).

Note

that

Gm)

= 0 for any local

ring

R

([M], III.4.10),

hence the scheme Y x x R is

isomorphic

to

Gm, R.

We have Pic

Gm,s -

Pic ,5’ for any normal

base scheme S. Now 0 which

implies Pic (Y

x x

R) =

0.

Therefore, RI f*Gm

= 0.

Recall that a

ring

of finite

type

over an excellent

ring (resp.

the

localization of an excellent

ring)

is excellent

( ~EGA4~ , 7.8.3).

Hence the local

ring Ox,x

is an excellent

ring.

Since k is of characteristic zero, Hironaka’s resolution of

singularities

holds for

Ox,x. By [EGA4],

7.9.5 and

18.8.17,

the

(7)

From

(2), (3)

and statement

(i)

of this lemma we obtain an exact sequence off-modules

([CS87a], Prop. 2.1.1,

see also

[SO1], 1.3):

The map

T -

Pic X in this exact sequence is called the

type

of the torsor

f :

Y - X. The torsor

f :

Y - X is called universal if its

type

is an

isomorphism.

In this case we have Pic Y = 0. When =

1~*

the

type

of the torsor

f :

Y - X is

injective

if and

only

if

~[V]*

=

k*.

In

particular,

this

property

holds for universal

torsors,

and

implies Bri

Y =

Bro

Y = Br k

by (1).

In this paper we define a relative smooth

compactification

of a torsor

under a torus as follows.

DEFINITION 1.2. - Let

f :

Y -~ X be a torsor under a torus. A

relative smooth

compactification

of

f :

Y - X is a

geometrically integral variety

Z endowed with a smooth and proper

morphism g :

Z - X with

connected

geometric fibres, together

with an

embedding

i : Y - Z such

that

i(Y)

is open and dense in

Z,

and

f =

g o i.

The existence of a relative smooth

compactification

in characteristic 0 follows from Hironaka’s theorem.

DEFINITION 1.3

[CS77].

- An exact sequence off-modules

such that

M, P, F

are free abelian groups of finite

rank,

is called a

flasque

resolution of M if the

following

conditions hold:

(8)

(a)

P is a

permutation r -module,

that

is,

it has a r-invariant Z-basis.

(b)

F is

flasque,

that

is,

the Tate

cohomology

group

F)

is

trivial for any

subgroup

G’ C

G,

where G is the

image

of the action of r

on F.

LEMMA 1.4. - Let X be a smooth

k-variety,

and let

f :

Y --~ X be a

torsor under a torus T of

type

T :

T -

Pic X.

Let g :

Z - X be a relative

smooth

compactification

of

f :

Y - X. Then

(i)

there is a natural

flasque

resolution

of T :

Proof - We prove

(i)

and

(ii) simultaneously.

Since Z is

regular

we

have an exact sequence of sheaves in the 6tale

topology

on Z:

The sheaf

DivzBy

is defined as the direct sum of

j,,Z, where j

is the

embedding

into Z of a closed

integral

subset of codimension 1 contained in

Z B Y,

for all such closed subsets of Z. The sequence

(5)

is exact since Z is

regular,

and hence the Weil divisors coincide with Cartier divisors

(see [M], II, Example 3.9).

Let

D+ (X) (resp. ’D+ (Z))

be the derived

category

of bounded below

complexes

of 6tale sheaves on X

(resp.

on

Z).

The derived functor

I~g* : D+ (Z)

-~

D+ (X ) gives

an exact

triangle

in

D+ (X) :

In

D+ (X)

there is also an exact

triangle

defined

by (3).

There is a natural

isomorphism

= The canonical

morphism

-1

is an

isomorphism

in our case

since g

is proper with connected fibres. These two

morphisms

define a

morphism

of exact

triangles

in

D+(X):

Let K be the function field

k(X), Ko - k(X).

We have the

following

identifications:

(9)

From

(6)

we obtain the

following

commutative

diagram

of r-modules with exact rows and

columns,

which is the definition of F:

The arrows in the

right

hand square are the natural maps. The

top

row

agrees with the exact sequence

(4);

the map bi~ - Pic X is the

type

T of the

torsor

f :

Y - X. We observed that

Gm,x - g*Gm,z.

This

implies

that

k[X]*

=

k[Z]*,

and that Pic X

injects

into Pic Z

(by

the

Leray spectral sequence).

Now it follows from the

diagram

that the map

T

-~

is

injective.

This proves the exactness of the sequence in

(i).

To

complete

the

proof

of

(i)

we now show that F is a

flasque

r-

module. Let k’c k be the smallest

(necessarily Galois)

extension of k such

that Gal

(k/k’)

acts

trivially

on the modules in

diagram (7).

Call G the

finite group Gal

(k’lk).

Note that the

generic

fibre

ZKo

smooth

compactification

of the

principal homogeneous

space

YKo

= Y x x

Ko

of the

torus T x x

Ko.

We

identify

r = Gal with Gal

(K/Ko ) .

Then

we have a canonical

isomorphism

of r-modules

T (see,

e.g.

[SOl],

Lemma

2.4.3).

This leads to an exact sequence of r-modules

The exact sequence

(8)

will not

change

if we

replace

the extension K =

k(X)

of

Ko

=

k(X) by

the extension L =

k(Y) of Lo

=

k(Y). Indeed,

since

(10)

Ko

is

algebraically

closed in

Lo

the Galois group Gal

(L/Lo )

is

canonically isomorphic

to r = Gal

(K/Ko ) .

On the other

hand,

we are

dealing

with r-

modules which are free abelian groups of finite

type

that do not

change

after

the passage from K to L. The

advantage

now is that Y has a canonical

Lo- point given by

the

embedding

of the

generic point

into Y. Hence the Galois module Pic

ZL

= Pic

ZK

is the Picard module of a smooth

compactification

of the torus

TL.

The Galois group r acts on

Pic ZL

via its

quotient

G.

By

a theorem of Voskresenskii

([V],

Thm.

4.6)

the G-module

Pic ZL

is

flasque.

The G-module Div is the direct sum of the module of ’vertical’

divisors

Div v,

that

is,

the divisors on Z that do not intersect the

generic

fibre

ZK,

and the module of ’horizontal’

divisors,

that

is,

the divisors on Z that are Zariski closures of the divisors on

Zx

with

support

in

Zx B

The G-module of horizontal divisors is thus identified with Div ZK B YK Restriction to the

generic

fibre defines a map of G-modules Pic Z - Pic This map is

surjective

because Z is

regular.

Its kernel

obviously

contains

Pic X,

hence it factors

through

a

surjective

map F --~

Pic Z K .

The

composition

of the map T -~ from

diagram (7)

with

the map --~ Div

(Zx) given by

the restriction of divisors to the

generic

fibre

Zx,

is none other than the second map from

(8). Indeed,

the

right

hand vertical arrow in

(6) gives

rise to a commutative

diagram

which is

precisely

what we need:

Putting everything together

we obtain an exact commutative

diagram

of

G-modules

(11)

(The

exactness of the bottom row follows

immediately

from the exactness

of the columns and the

top row).

Since

Div v

is a

permutation module,

it is

flasque by Shapiro’s

lemma. Now the bottom line of the

diagram

shows that

F is

flasque (in fact,

an extension of a

flasque

module

by

a

permutation

module is

split).

This establishes

(i).

In

particular,

we have

~I - ~ (C, F) -

0 for any

cyclic subgroup

C C G.

By

the

periodicity

of Tate

cohomology

of

cyclic

groups we obtain

H 1 (C, F)

= 0 for any

cyclic subgroup

C C G.

We now go back to

diagram (7).

The

just

established

property

of

F,

and the fact that

DivZBY(Z)

is a

permutation G-module,

and hence

enable us to conclude

that

coincides with the

injective image

of in

H2(1~, T) (cf. [CS87b],

the

proof

of

Prop.

9.5

(ii)).

The

statement

(ii)

now follows from

diagram (7).

D

Remark 1.5. - We have associated a canonical

flasque

resolution

of

T

to any relative smooth

compactification

of an X-torsor under T.

This

generalizes

the construction due to Voskresenskii and Colliot-Thelene- Sansuc

(see [V], [CS77])

when the X is the

spectrum

of a field.

1.2. The

pull-back

of the Brauer group to a torsor.

In connection with arithmetic

applications

of torsors under tori we are interested in

describing

the cokernel of the map

f * :

Br X ---+

Br Y,

especially

when Y is a universal torsor.

Arguably

a more essential

question

(12)

is the

computation

of the cokernel of the map

f nr : Brnr

-

Brnr

Colliot-Thélène and Sansuc

pointed

out that when 1~ is

algebraically

closed the map

f nr

is an

isomorphism. (This

follows from the fact that in this situation Y is

birationally equivalent

to the

product

Our first result concerns the cokernel of

f * :

Br X - Br Y over an

algebraically

closed field.

THEOREM 1.6. - Let k be an

algebraically

closed field of charac- teristic 0. Let X be a smooth and

geometrically integral variety

over k

such that - k* and Pic X is a free abelian group of finite

type.

Let

f :

Y --~ X be a torsor under a k-torus such that

k[Y]*

= k* and Pic Y is

a free abelian group of finite

type (for example,

a universal

torsor).

Then

the map

f * :

Br X - Br Y is an

isomorphism.

Note that after

[CS87a]

the

injectivity

of

f * :

Br X - Br Y is known to be true

(over

an

algebraically

closed

field)

for any torsor under a

torus,

but the

surjectivity

does not hold in

general.

A well-known

example

is the

torsor under but Br

Proof of Theorem 1.6. - Let X be a smooth and

geometrically

connected

variety

over an

algebraically

closed field

k,

such that

k[X]*

= k*

and Pic X is torsion-free. Let

f :

Y --4 X be a torsor under T. We denote its

type by

Ty. The

assumptions

of Theorem 1.6 are

equivalent

to the condition that TY is

injective

and

Ty (T)

is a

primitive

sublattice of Pic X.

Let us choose a Z-basis in

T ~ Z’~,

and factor

f :

Y - X into a

composition

of torsors under

where each

Yi

is a

k-variety.

Then we have

H° (Y2, Gm)

= k*. It is also

clear that

Pic Yi

is a torsion free abelian group of finite

type.

It is therefore

enough

to prove Theorem 1.6 when T =

Gm.

We assume this from now on.

The

spectral

sequence

(2)

and Lemma 1.1

imply

that BrY =

H2 (X, f * Gm ) .

Now the exact sequence

(3) together

with the remark that follows

it,

and the fact that Br Y is a torsion group lead to an exact se-

quence

(13)

now follows from

(10)

that the map

f * :

Br X - Br Y is an

isomorphism.

Theorem 1.6 is

proved.

D

Over

general

fields the situation is more

complicated.

THEOREM 1.7. - Let k be a field of characteristic 0 such that

H3 (1~, I~* )

= 0. Let X be a smooth and

geometrically integral k-variety

such that

I~~X~* - k*

and Pic X is a free abelian group of finite

type.

(a)

Let

f :

Y - X be a torsor under a k-torus

T,

such that

I~ ~Y~ *

k’

and Pic Y is a free abelian group of finite

type (for example,

a universal

torsor).

Let T :

T

-~ Pic X be the

type

of

f :

Y - X. Let

P,

be the

intersection of the

image

of

(Br X ) r -~ H2 (I~, Pic X )

with

T*(H2(k, T ) )

in

Then there is an exact sequence of abelian groups

(b)

Let

f :

Y - X be a universal torsor. Then there is a natural

isomorphism

= such that the canonical map Br X -

(Br X ) r

identifies with the map Br X -

Br Y/Br k

defined

by f * .

Note

that,

in the

assumptions

of this

theorem,

the

surjectivity

of the

map

f * :

Br X - Br Y for a universal torsor

f :

Y - X is

equivalent

to

the

surjectivity

of the map Br X -

Proof. - The natural map of r-modules

f* :

Br X ~ Br Y is

an

isomorphism by

Theorem 1.6. Hence

isomorphism.

is also an

We now

exploit

the

functoriality

of the Hochschild-Serre

spectral

sequence

with

respect

to the

morphism f :

Y - X. In view of our

assumption

H3 (1~, l~* )

= 0 this

spectral

sequence

gives

rise to the

following

commutative

(14)

diagram

with exact rows:

The statement

(a)

now follows from this

diagram

combined with exact sequence

(4).

To prove

(b)

we observe that for a universal X-torsor Y one has Pic Y = 0. From the

diagram

we obtain

Br Y /Br 1 Y == (Br Y)r,

and

the result follows. Theorem 1.7 is

proved.

D

For the sake of

completeness

we record a

proposition

which can

be

proved along

the same lines but without recourse to the condition

H3 (k, k*)

== 0. It

improves

a similar earlier result of Colliot- Thélène and Sansuc valid for smooth

compactifications

of universal torsors over

projective

varieties.

PROPOSITION 1.8. - Let k be a field of characteristic 0. Let X be a

smooth and

geometrically integral k-variety

such that

1~ ~X ~ * = 1~* ,

Pic X is

a free abelian group of finite

type,

and

(Br

= 0. Then for any universal X-torsor Y we have Br Y = Br k.

The

following theorem,

in which the torsor is

replaced by

a relative

smooth

compactification,

seems to be more useful in

applications.

THEOREM 1.9. - We

keep

the notation and conditions of Theorem 1.7

(a).

Let

QT

be the intersection of the

image

of

(Br X )r

-

H2 (k,

Pic

X )

with T*

(EL12 (k, T))

in

H2 (k,

Pic

X).

Then for any relative smooth

compact-

ification Z - X of Y --4 X there is an exact sequence of abelian groups

Proof. The

beginning

of the

proof

is similar to the

proof

of

Theorem 1.7.

Let g :

Z ~ X be any relative smooth

compactification

of

f :

Y - X. This time we use the

functoriality

of the Hochschild-Serre

spectral

sequence

with

respect

to the

morphism g :

Z - X.

The natural map

g* : (Br X )r -~ (Br Z)r

is an

isomorphism. Indeed,

by [G]

the map i* : : Br Z - Br Y is

injective (since

Y C Z is an open

(15)

The

proof

is now

completed by

an easy

diagram

chase and an

application

of Lemma 1.4. D

For a proper

variety

X we

obtain,

as a

corollary,

a result about the cokernel of the map Br.

COROLLARY 1.10. - In the

assumptions

of Theorem 1.7

(a)

assume

further that X is proper. Then we have an exact sequence

where

The

corollary

follows from Theorem 1.9 because in this case Z is a

smooth and proper model of

Y,

hence Br Z =

Brnr by [G],

6.2.

Since

0,

we have

Brk,

whence the

following corollary:

COROLLARY l.ll. In the

assumptions

of Theorem 1.7

(a)

as-

sume further that X is proper and Y is a universal X-torsor. Then

Brnr

=

Br X/Bri

X whenever Pic

X) -

0.

In the case of a number field k we have a more

precise

result.

PROPOSITION 1.12. - Let X be a proper, smooth and

geometrically

integral variety

over a number field k such that Pic X is a free abelian group of finite

type.

Let Y be a universal X-torsor. Then the

quotient

of

(16)

the unramified Brauer group

Brnr (k(Y)/k) by

the

image

of Br X is the

intersection

where and v ranges over

the

places

v of k such that the r-module Pic X is unramified at v

(the

inertia group at v acts

trivially).

Proof. The

functoriality

of the Hochschild-Serre

spectral

sequence with

respect

to restriction from k to

kv gives

rise to a commutative

diagram

with

injective

horizontal arrows:

It follows from the Tchebotarev

density

theorem that Pic

X)

is

the intersection of the kernels of the restriction maps

H2 ( k, Pic X) ---+

H2(kv,PicXv)

for v such that the r-module Pic X is unramified at v.

Now the

diagram

shows that

Qid

is

given by

the

displayed

formula of the

proposition.

On the other

hand,

for a universal X-torsor Y - X we have

0,

whence the

following

exact sequence:

The

proposition

follows. D

It would be

interesting

to

give

an

example

of a

variety

X

satisfying

the

assumptions

of this

proposition

such that

Brnr

contains an element that does not come from Br X.

Remark 1.13. - If

Y(I~) ~ 0,

then the existence of a section of the map

H3 (1~,1~* )

-

H3 (y, Gm)

makes the

assumption H3(k, k*) ==

0 in

Theorem 1.7

superfluous.

The same is true for Theorem 1.9 if

Z(1~) ~ 0.

2.

Arithmetic

of

vertical

torsors.

In this section we cross the fibration method of

[H94]

and

[H97]

with

the descent method of

~590~, [S96]

and

[CSOO].

2.1. Main theorem.

Let 1~ be a field of characteristic

0,

and X be a

geometrically integral

k-variety equipped

with a

surjective morphism f

to

P~ . Let q

be the

generic

(17)

over the residue field

k(P)

of P.

LEMMA 2.1. Let X be a

smooth, projective

and

geometrically integral k-variety.

Let

f :

X - be a

morphism

with

geometrically integral generic

fibre which contains a Then there exists a dense open subset U C X such that

(a)

the restriction of

f

to U is a smooth

surjective morphism

with

integral

closed

fibres ;

(b)

if for a closed

point

P E

pl

the fibre

Xp

is

split,

then the fibre

Up

is

geometrically integral;

(c)

the

generic

fibre

UK

coincides with

XK.

The same conclusion holds under a weaker

assumption

that

f :

X -

P~ locally

in etale

topology

has a section at every closed

point P

G

P~.

Equivalently,

every closed fibre of

f

contains an irreducible

component

of

multiplicity

1.

Proof. For any closed

point

P of

P~

the fibre

X p

contains an irreducible

component

of

multiplicity

1.

Indeed, XK

has a

k(71)-point

and

is proper, thus

f

has a

k-section.

Since X is

regular,

the intersection

point

of this section with

X p

must be

regular

on

Xp.

In

particular,

this

point belongs

to an irreducible

component

of

multiplicity

1.

For every closed

point

P choose an irreducible

component Ep

C

Xp

of

multiplicity

1. Whenever there is a

geometrically

irreducible

component

we choose it as

Ep.

Let

Fp

be the union of all the

remaining components

of

X p ,

and let

G p

be the

singular

locus of

Ep.

The

complement

U to the

(finite)

union of the

Fp

and the

Gp

for all the closed

points

P E

P~

has

the

required properties.

D

Let

(18)

A r-module M such that M =

MU

for a dense open subset U C X

satisfying

the conditions of Lemma 2.1 will be called a vertical module associated to

f.

A U-torsor of

type Mu -

Pic U is called vertical. It is easy to see that

MU

is a torsion-free abelian group

([S96], Prop. 3.2.3).

THEOREM 2.2. - Let X be a

smooth, projective

and

geometrically integral variety

over a number field

k,

and let

f : X - P~

be a

dominant

morphism

with

geometrically integral generic

fibre.

Suppose

that

the

following

conditions hold:

(2) PicXj(

is

torsion-free,

Br

XK

is

finite,

and

XK

has a

(3)

either Br

Xk(,) - 0,

or there exists a vertical module M associated to

f

such that

M)

= 0.

If the Manin obstruction is the

only

obstruction to the Hasse

principle

or to weak

approximation

for the k-fibres of

f

in a dense open subset of

P’,

then the same

property

holds for X.

This theorem should be

compared

with Thm. A of

[CSOO].

Under

additional

cohomological assumptions

we

replace

the arithmetic

hypothesis

that the k-fibres of

f satisfy

the Hasse

principle

or weak

approximation by

a weaker condition. The

hypothesis (2)

of the theorem is satisfied when

XK

is a

rationally

connected

variety.

In this case the existence of a

on

XK

follows from a recent theorem of

Graber,

Harris and Starr

[GHS].

Let us describe the idea of

proof

of Theorem 2.2. Recall that

is the set of adelic

points orthogonal

to Br U with

respect

to the Brauer- Manin

pairing.

As in

[S96]

and

[CSOO]

we can lift an adelic

point

of to an adelic

point

on some vertical torsor Y - U. The

difficulty pointed

out in

[H97], 4.5,

is that in our situation there is no reason

why

Y should

satisfy

the Hasse

principle:

the

hypothesis

that the

Manin obstruction to the Hasse

principle

is the

only

one for the k-fibres of

f

is too weak for this.

However, using Proposition

2.4 below and the results of Section 1 we show that our condition

(3) implies

that

IPl

lifts to an adelic

point

E where Z is a relative smooth

compactification

of Y. On the other

hand,

we show that on Z the Manin obstruction to the Hasse

principle

is the

only

one: this is a consequence of the

hypothesis

2 and the results of

[H97] (which replace

Thm.

2.1 of

[S96]).

Therefore the

assumption 0 implies

that

Y(k)

is not

empty,

hence

U(k)

is

non-empty.

As is dense in

(19)

PROPOSITION 2.4. - Let X be a smooth and

geometrically integral variety

over a number field

k,

such that

Gm ) = l~*

and Pic X is an

abelian group of finite

type.

Let T be a

k-torus,

and T be an

injection

of r -

modules

T

--+ Pic X.

Suppose that ~ Mv ~

E . Then there exists

a torsor

f :

Y - X under T of

type

T, and a

point

E

such that

Mv

=

f (Pv )

for all

places

v.

Proof. Let S be the

k-group

of

multiplicative type

dual to the h- module

Pic X,

5’ 2013~ T be the

surjective k-homomorphism

dual to T, and

S’1

be its kernel. We denote

by

Id the canonical

isomorphism S -

Pic X.

Under our

assumptions

the main theorem of the descent

theory ([CS87a], [SOl], 6.1.2)

states that an adelic

point

on X

orthogonal

to

Br, X

with

respect

to the Brauer-Manin

pairing

can be lifted to an adelic

point

on some universal torsor Z ---&#x3E; X of

type

Id. Then the

quotient

Y = is

an X-torsor under T of

type

T

(this trivially

follows from the

functoriality

of

type

with

respect

to the structure

group).

Let us prove that Z - Y is a universal torsor over Y. Since Z is a universal X-torsor and

= 1~*,

we see from

(4)

that

HO(Z,Gm)

=

k*

and

Pic Z

= 0. The same exact sequence

(4)

considered

for the torsor Z - Y now shows that T - Pic Y is an

isomorphism,

hence

Z is a universal Y-torsor.

Now let be the

image

of on Y. The inverse main theorem

of the descent

theory ([S01], 6.1.2)

states that the

image

of an adelic

point

on a universal torsor is

orthogonal

to

Bri

Y. This is

exactly

what we wanted

to prove. D

Note that this statement is non-trivial in the

following

sense: in

general

the natural map

Bri

X -

Bri

Y is not

surjective.

We shall need the

following slight

refinement of

[H97],

Thm. 3.2.1.

(20)

PROPOSITION 2.5. - Let F be a closed subset

of A’

of codimension at least

2,

L =

k(A’).

Let V be a

quasi-projective k-variety equipped

with

a

surjective morphism

p : V ~

A~ B

F with

split

fibres and

geometrically integral generic

fibre

VL.

Let

XL

be a smooth and

projective

model

of VL.

Assume the

following

conditions:

(a)

The restriction

of p

to the

generic point

of a

sufhciently general

line D in

A’

has a smooth

k(D)-point.

(b)

Br

XL is finite,

and Pic

(XL) torsion-free,

where L is an

algebraic

closure

of L,

and

XL = XL

X L L.

(c)

There exists a

non-empty

open subset Q

of An

such that for each 9 e

SZ(1~),

the Manin obstruction to the Hasse

principle (resp.

to weak

approximation)

is the

only

one for smooth and

projective

models of the fibre

Ve .

Then we have the

following

statements:

(i)

The Manin obstruction to the Hasse

principle (resp.

to

weak

approximation)

is the

only

one for a smooth and

projective

model V of

V.

(ii)

Assume further that

0.

Let Q’ be any

non-empty

open subset

of An.

Then there exists a

k-point 9

of 0’ such that a smooth and

projective

model of the fibre

Ve

has a

k-point.

In

particular, if VL

is smooth and

projective,

and

0,

then

p(V(k))

is Zariski dense in

An

Proof -

(i)

is

precisely [H97],

Thm. 3.2.1. We prove

(ii) using

the

same induction

argument.

As in the

proof

of

Prop

3.1.1 of

[H97]

we embed

V into a

projective k-variety Vi,

and consider the closure V’ of the

graph of p in VI xk (A~ B F).

We obtain a

projective morphism p’ :

V’ -~

A’ B F

whose

generic

fibre

VL

is

birationally equivalent

to

VL.

In

particular, p’

satisfies the

assumptions (a), (b)

and

(c)

of the

proposition.

The fibres of

p’

are

split

because a

k-variety containing

a

split non-empty

open subset is

split.

In the

case_n -

1 it follows

immediately

from

[H97], proof

of

Prop. 3.1.1,

that if

0,

then any

non-empty

open subset of

A)

contains a

k-point

such that the

corresponding

fibre has smooth

k-points.

This proves

(ii)

for n = 1

by

Nishimura’s lemma.

Assume that

(ii)

is

proved

for any

integer

less than n. As in

[H97], proof

of Thm.

3.2.1,

we can find a

k-point Mo

in

An

such that the k-

morphism

obtained

by composing p’

with the

(21)

variety satisfying

the

following properties:

the

generic

fibre of the restriction

pD :

is smooth

and has a

k(D)-point;

o the

geometric generic

fibre

of pD

has finite Brauer group and torsion- free Picard group.

Since all the fibres

of pD

are

split (because

the same

property

holds for

p’),

we can

apply

the case n = 1 of

Proposition

2.5

(ii)

to

p’n. (Recall

that

0.)

We obtain that each

non-empty

open subset of D contains

a k-rational

point 0

such that a smooth and

projective

model of the fibre

p’-1 (B)

has a

k-point.

This proves

(ii)

for

p’,

hence for p

by

Nishimura’s

lemma. 0

Remark 2.6. -

Although

we are

only

interested in the conclusion of

Proposition

2.5 when the

generic

fibre is smooth and

projective,

we need

a more

general

statement for the induction to work.

Proof of Theorem 2.2. - The case 1 is known

([H97], Prop.

3.1.1,

here condition

(3)

is not

required).

We now assume that

rk f

= 2.

If B =

0,

then we take any U

satisfying

the conditions of Lemma 2.1. In this case Br U is trivial since it is a

subgroup

of Br

by

Grothendieck’s theorem.

If 0 we choose a vertical module M associated to

f

with

M)

=

0,

which is

possible by

condition

(3).

In this case we define U

to be the dense open subset of X such that M =

Mu.

We observe that

k [U] *

= k*.

Indeed,

the

generic

fibre of

U -~

PI

is proper and

geometrically integral,

hence every

regular

function

on comes from a rational function on

P~. However,

if such a function

is

regular

on

U,

then it must be constant due to the

surjectivity

of the

morphism

U -

Pl.

k

*

Références

Documents relatifs

The “Live@Knowledge 5.0” computer program will be presented to all those people who shared their data and knowledge to build knowledge needed to solve historical problems.. It is

We study the deformed Hermitian-Yang-Mills (dHYM) equa- tion, which is mirror to the special Lagrangian equation, from the vari- ational point of view via an infinite dimensional

We will use the idea of Bakry and Emery [1] to give a notion of a ”lower Ricci curvature bound” on graphs and to prove that Ricci curvature on Locally finite graphs is always

We prove global existence and asymptotic convergence of the solution for the p-pseudoharmonic map heat ‡ow, provided that the sectional curvature of the target manifold N

First introduced by Faddeev and Kashaev [7, 9], the quantum dilogarithm G b (x) and its variants S b (x) and g b (x) play a crucial role in the study of positive representations

James presented a counterexample to show that a bounded convex set with James’ property is not necessarily weakly compact even if it is the closed unit ball of a normed space.. We

In this paper, we propose a nonlocal model for linear steady Stokes equation with no-slip boundary condition.. The main idea is to use volume constraint to enforce the no-slip

We prove that arbitrary actions of compact groups on finite dimensional C ∗ -algebras are equivariantly semiprojec- tive, that quasifree actions of compact groups on the Cuntz